---
title: Synth Cue Restoration Studio
emoji: 🎛️
colorFrom: purple
colorTo: blue
sdk: gradio
sdk_version: 6.19.0
app_file: app.py
pinned: false
---

# Synth Cue Restoration Studio

A Gradio tool for restoring degraded analog-synth score cues transferred
from flawed master tapes. Upload a degraded audio file and, optionally,
a clean reference clip to guide tonal (EQ) matching.

## What it does

- **Noise reduction** — spectral-gating denoise for tape hiss / transfer noise
- **Declick / dropout repair** — detects and interpolates across impulsive
  clicks, pops, and short dropouts, using a local (not global) threshold
  so it doesn't fire on dense musical passages
- **Wow & flutter smoothing** — gently corrects slow pitch/speed drift from
  tape stretch or warp
- **High-frequency restoration** — a harmonic exciter that regenerates
  plausible top-end lost to bandwidth-limited transfers
- **Reference spectral + stereo-width matching** — provide a clean
  reference clip and the tool measures its tonal balance and stereo width
  live, then nudges your restored cue toward those measured values

This is a general-purpose DSP restoration pipeline. It processes whatever
audio you upload — it does not generate or reproduce any copyrighted
composition on its own.

### Important: always use a matching reference, not a generic one

Analysis across multiple clean cues from this score showed no single
consistent "target" tone — stereo width ranges from near-mono to very
wide/decorrelated, and brightness ranges from ~6kHz to ~17.6kHz rolloff
depending on the specific cue. There is no universal profile to aim for.
**Always upload the clean counterpart of the exact cue you're restoring**,
not an unrelated reference clip, or the spectral/width matching will
nudge the audio toward the wrong target.

### Presets

Three one-click presets seed the sliders based on measured degradation
patterns, not tonal target:
- **Broadband Noise Cue** — bandwidth already close to intact relative to
  its own reference, main issue is hiss
- **Bandwidth-Limited Cue** — real high-frequency loss and a narrower
  stereo image relative to its own reference
- **Heavy Click / Worn Duplication** — click rate several times higher
  than typical, often with reduced dynamic range

## Running locally

```bash
pip install -r requirements.txt
python app.py
```

## Deploying to Hugging Face Spaces

1. Create a new Space at huggingface.co/new-space, SDK = **Gradio**.
2. Copy `app.py`, `requirements.txt`, and this `README.md` into the repo
   (or `git clone` the Space repo and drop these files in).
3. `git add . && git commit -m "Initial restoration studio" && git push`
4. The Space will build automatically and give you a live URL.